Embodied Referring Expression Comprehension on CAESAR-PRO
42.91IoU (0.25)CLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLIPResidual Module Configuration=MuRes(V)2025.12 | 42.91 | 11.91 | |
| Dual-EncoderResidual Module Configuration=Vanilla Residual2025.12 | 42.61 | 11.61 | |
| Dual-EncoderResidual Module Configuration=Without Residual2025.12 | 42.52 | 12.14 | |
| CLIPResidual Module Configuration=MuRes(L)2025.12 | 39.56 | 10.85 | |
| CLIPResidual Module Configuration=Vanilla Residual2025.12 | 39.43 | 10.83 | |
| CLIPResidual Module Configuration=MuRes(V+L)2025.12 | 39.06 | 10.46 | |
| Dual-EncoderResidual Module Configuration=MuRes(L)2025.12 | 37.97 | 10.32 | |
| CLIPResidual Module Configuration=Without Residual2025.12 | 37.92 | 9.82 | |
| Dual-EncoderResidual Module Configuration=MuRes(V+L)2025.12 | 37.72 | 11.5 | |
| Dual-EncoderResidual Module Configuration=MuRes(V)2025.12 | 36.72 | 8.51 | |
| ViLTResidual Module Configuration=MuRes(V+L)2025.12 | 28.52 | 8.04 | |
| ViLTResidual Module Configuration=Without Residual2025.12 | 27.96 | 8.73 | |
| ViLTResidual Module Configuration=Vanilla Residual2025.12 | 25.67 | 8.06 |